First, thanks to everyone who's currently involved in this project; you're making history, guys! 💜
nostr:npub1wmr34t36fy03m8hvgl96zl3znndyzyaqhwmwdtshwmtkg03fetaqhjg240 while I deeply empathize with the stress and understand the project status, I can't be ignorant about p2p. As far as I see, there's currently no sign of any (working) p2p media content delivery in the project, is it right? Is it a strategic choice?
IMO p2p should be prioritized over the backend challenges in this case. I believe it's the main reason why PeerTube succeed at scaling: a single shitty VPS can handle 1k+ concurrent viewers, and I'm not aware whether the upper bound is even known, but I'm sure it can handle more with the right WebTorrent tracker and manual configuration https://joinpeertube.org/news/stress-test-2023 And yeah, I know, PeerTube solves a different problem: the long videos.
I've got a YouTube channel with 30k+ subscribers, but I no longer post full videos there because of much earlier signs of YouTube degradation (no dislike counter anymore; no way to completely disable the crappy ads, even by the channel owner, etc.). So I run my own video hosting for my subscribers, using the same technologies as PeerTube does, which is basically https://github.com/Novage/p2p-media-loader
Not sure whether this has been researched, but anyway, in two words: p2p-media-loader is basically HLS + WebTorrent wrapper, WebTorrent is the Torrent on top of WebRTC. There are WebRTC-related NIP proposals already btw.
My setup is different than the PeerTube itself though: most viewers receive av1 + opus (h264 + aac as fallback, mostly for iOS), small media files (instead of a single file), https://github.com/vidstack/player as player, https://github.com/greatest-ape/aquatic as tracker
The only crappy cheap (5 EUR/month) VPS instance I have (with a literally failing SSD) but with fast enough outgoing bandwidth (around 1 Gbps) worked surprisingly well. I've got no idea how many concurrent viewers I had exactly, but the poll statistics from a small subset of subscribers seemed positive enough to me.
I also may give a viewer proxified media files over Cloudflare, but only as a fallback, in case such viewer refreshes the page (it's a dirty workaround for slow warmup/seek, primarily designed for those, who watch from long distance from the server, where latency becomes, say, >300ms).
My guess is that in the end of the day, AI enshittification will be resolved in a different elegant way (with simpler things like WoT), due to the Turing test passing tendency: if a user at some point won't be able to distinguish AI vs human content, then why would an algorithm be able to do that? How would another human be able to verify that algorithm works? To what degree should content be "real"? A computer scientist that demonstrates video about AI is prohibited? There will be lots of false positives/exceptions and publishing human content will become as painful as passing CAPTCHA today. But it's a separate philosophical discussion; I'm not denying that detecting AI-generated media content is a necessary measure right now; I'm only advocating that it's not a sustainable strategy in the long run (or at least it's not enough).
Anyway, just in case, I'm glad to answer anything regarding P2P content delivery here in this thread; I'll answer whatever I'm aware of, if it's useful in any way.